chronax.loss

Loss functions for the Chronax FEDformer model (self-contained, pure JAX). The objective functions optimised during training and used for validation.

---

mae

chronax.loss.mae

Mean Absolute Error; masked mean when mask is given (NF padder).

Parameter	Type	Default	Description
pred	jnp.ndarray	-	(undocumented)
target	jnp.ndarray	-	(undocumented)
mask	Optional[jnp.ndarray]	None	(undocumented)

mse

chronax.loss.mse

Mean Squared Error; masked mean when mask is given.

Parameter	Type	Default	Description
pred	jnp.ndarray	-	(undocumented)
target	jnp.ndarray	-	(undocumented)
mask	Optional[jnp.ndarray]	None	(undocumented)

huber

chronax.loss.huber

Huber loss with delta = 1.0 (smooth L1); masked mean when mask is given.

Quadratic for small residuals (|r| <= 1) and linear beyond, giving the smooth gradients of MSE near zero with the outlier-robustness of MAE in the tails.

Parameter	Type	Default	Description
pred	jnp.ndarray	-	(undocumented)
target	jnp.ndarray	-	(undocumented)
mask	Optional[jnp.ndarray]	None	(undocumented)

LOSSES

chronax.loss.LOSSES

Registry of built-in window losses, addressable by name.

Type: Mapping[str, LossFn]

resolve

chronax.loss.resolve

Return a loss callable from a registry key or pass a callable through.

Parameter	Type	Default	Description
loss	str \| LossFn	-	Either a registry key ("mae" / "mse" / "huber") or any (pred, target) -> scalar callable.

Returns: LossFn (the resolved loss callable). Raises: ValueError (If loss is an unknown string key.)

masked_mae

chronax.loss.masked_mae

Mean Absolute Error with multiplicative masking.

Parameter	Type	Default	Description
y	jnp.ndarray	-	Target values [B, H, 1].
y_hat	jnp.ndarray	-	Predictions [B, H, 1].
mask	Optional[jnp.ndarray]	None	Optional [B, H, 1] mask (1 = include, 0 = ignore).
horizon_weight	Optional[jnp.ndarray]	None	Optional [H] per-step weights.

masked_mse

chronax.loss.masked_mse

Mean Squared Error with multiplicative masking (see :func:masked_mae).

Parameter	Type	Default	Description
y	jnp.ndarray	-	(undocumented)
y_hat	jnp.ndarray	-	(undocumented)
mask	Optional[jnp.ndarray]	None	(undocumented)
horizon_weight	Optional[jnp.ndarray]	None	(undocumented)
```